What problem does it solve? Autonomous agents need a disciplined, auditable process for shipping new features without bypassing human governance or breaking the build. This Skill enforces a single-pass feature workflow—read request, tier the risk, plan, implement, test, commit, and delegate push/CI—so feature work stays within an approved autonomy ceiling. ## Core Features & Use Cases - Risk-tier gating: Classifies work against a 1-10 tier ladder and routes anything above the autonomous ceiling (tier 3) to plan-only mode requiring human approval. - Non-waivable fix-bar: Requires tests, docs, atomic commits, and no scope creep before any commit lands; failures trigger revert and escalation. - Peer delegation: Hands push and CI watch to the iris peer agent, with fix-forward loops (max 2 attempts) and automatic tier demotion after triggered reverts. - Use Case: A user says "build a new MCP tool for log search"—the agent fetches the request, tiers it, writes a structured plan to feature_plans.md, implements with tests and docs in atomic commits, and delegates push plus CI monitoring. ## Quick Start Ask the agent to build a specific feature, for example: "felix, implement a health-check endpoint for the harness scheduler."
